Applied Mathematical Sciences, Vol. 6, 2012, no.

106, 5253 - 5261

Bayesian Estimation of Generalized Half Logistic

Distribution Under Progressive Type-II Censored
Reza Azimi1 , Farhad Yaghmaei2 , and Manoochehr Babanezhad2
1

Department of Statistics, Faculty of Sciences

Golestan University, Gorgan, Golestan, Iran
azimireza1365@gmail.com

Department of Statistics, Faculty of Sciences

Golestan University, Gorgan, Golestan, Iran
f yaghmaei@yahoo.com, m.babanezhad@gu.ac.ir
Abstract
In this paper, given a progressively type II censored sample from
a generalized half logistic distribution, the Bayesian estimators are obtained under asymmetric loss functions such as LINEX loss function,
Precautionary loss function, Entropy loss function for the parameter,
Reliability and cumulative hazard functions. Comparisons of these estimators are made through simulation studies.

Mathematics Subject Classification: 62B15

Keywords: Asymmetric loss functions, Reliability function , Hazard function, Progressive Type-II censoring
Corresponding author : Manoochehr Babanezhad, e-mail: m.babanezhad@gu.ac.ir

Introduction

Consider two random variables X and Z where X = |Z|. Then X is called

the folded or half logistic random variable and Z is called the standard logistic random variable. The density function of X is a monotonic decreasing
function and has an increasing hazard rate. The generalized versions of half
logistic distribution namely Type-I and Type-II were considered along with
point estimation of scale parameters and estimation of stress strength reliability based on complete sample by [1]. The cumulative distribution function
(cdf), and the probability density function (pdf) of the generalized half logistic

5254

R. Azimi et al

distribution with parameter > 0 are respectively as follows,



2ex
F (x|) = 1
1 + ex

, x>0

(1)

(2ex )
(2)
f (x|) =
(1 + ex )+1
The reliability function R(t) and the cumulative hazard function (chf) h(t) at
mission time t are given respectively as follows,


2et
R(t) =
1 + et

2et
H(t) = log
1 + et
Progressive Type-II censored sampling is an important method of obtaining
data in lifetime studies. A recent account on progressive censoring schemes
can be obtained in the monograph by [2] or in the excellent review article by
[3]. Suppose that n independent items are put on a test and that the lifetime
distribution of each item is given by the probability density function of (2).
The ordered m-failures are observed under the type-II progressively censoring
m

plan (R, ..., Rm ) where each Ri 0 and
Rj + m = n. If the ordered mj=1

failures are denoted by x(1) < x(2) < ... < x(m) , then the likelihood function
based on the observed sample x(1) < x(2) < ... < x(m) (for convenience notation
are denoted by x1 < x2 < ... < xm ) is
L() = c

m

i=1

f (xi |)[1 F (xi |)]Ri

(3)

where c = n(n 1 R1 )...(n R1 ... Rm1 m + 1). Substituting (1), (2)

in (3), The latter function can be obtained as follows,
m


2exi
L() = c
1 + exI
i=1

 

2exi
1 + exi

Ri

(1 + exi )1

m exp {T (xi )}
where

m


2exi
T (xi ) = (Ri + 1) ln
1 + exi
i=1

(4)


Therefore MLE for can be obtained as the following form

m
=
T (xi )

(5)

and that the MLE of reliability function, R(t), and chf, H(t) can be obtained

= log 2ett .
= 2ett , and H
by R
1+e

1+e

5255

Bayesian estimation of generalized half logistic distribution

Bayesian Estimation

We now derive the Bayesian estimators for the parameters , reliability, and
hazard functions of the generalized half logistic distribution based on the progressive Type-II censored data. To do so, we consider family of prior densities
as the following form
ba a1 b
() =
e .
(a)
By combining (4) and the latter prior density function, we can obtain posterior
density of as the following form,
(b T (xi ))m+a m+a1 (bT (xi ))
(|x) =
e
.

(m + a)
Substituting =

log s
t
log 1+et

(6)

into (6) , we can obtain the posterior density function

2e

of s = R(t) where 0 < s < 1, as the follows,



(b T (xi ))m+a
1 + et
(s|X) =
log
(m + a)
2et

(m+a)

( log s)

m+a1

bT (xi )
1+et
2et

log

To derive the Bayesian estimators of the cumulative hazard function H(t) =

log R(t), we rst obtain the posterior density function of h = H(t), by
h
into (6), as the following form
substituting =
1+et
log(

2et




(h|X) =


t (m+a) h

(b T (xi ))m+a m+a1

1+e
h
log
(m + a)
2et


bT (xi )
1+et
log
2et

where h > 0. The Bayesian estimator of under the squared-error loss is the
posterior mean, that is
S = E(|X) =

m+a
b T (xi )

(7)

Under squared-error loss function, the Bayesian estimators of R(t) and H(t)
S and H
S are given by
denoted by R

m+a

b T (xi )
S =

R
2et
b T (xi ) log 1+e
t


S = S
H

(8)

1 + et
log
.
2et

The LINEX loss function for can be expressed as the following proportional
L() exp(c) c 1;

c = 0

5256

R. Azimi et al

where = ( ) and is an estimate of . The Bayes estimator of ,

denoted by L under the LINEX loss function is then obtained as follows
1
L = ln E [exp(c)]
c
By calculating the latter, we will obtain,


b T (xi )
m+a
log
.
L =
c
b + c T (xi )

(9)

To estimate Bayesian estimations for R(t) and H(t) under the LINEX loss
L and
function, we can expand ecs in Taylor series. Then we can obtain R

HL ,

m+a

m+a

b T (xi )
c2

+
2et
2! b T (xi ) 2 log 1+e
t
(10)


b

T
(x
)
m
+
a
i
L =
H
.
log
t
c
b T (xi ) + c log 1+e
t
2e

b T (xi )
L 1 log 1 c

R
2et
c
b T (xi ) log 1+e
t

In many practical situations, it appears to be more realistic to express the loss

in terms of the ratio . In this case, [4] points out that a useful asymmetric
loss function is the Entropy loss function,
L() [ P P loge () 1].

where = and its minimum occurs at

in [5] and [6], in the original form having
b > 0. Therefore, the Bayes estimator for
with P = 1 (denoted by E ) is given by,


= . The latter has been used

P = 1; L() = b[ logE () 1],
under the Entropy loss function

1
E = E( |X)

1

Then, we can obtain

m+a1
(11)
E =
b T (xi )
E ) and hazard function H(t) (deThe reliability function R(t) (denoted by R
E ) under Entropy loss function can also be as the following form
noted by H

(m+a)

b T (xi )
E =

R
2et
b T (xi ) + log 1+e
t


E = E log 1 + e
H
2et

(12)

Bayesian estimation of generalized half logistic distribution

5257

Precautionary loss function which is asymmetric, was introduced by [7],

) =
L(,

( )2
.

The Bayes estimator under the latter loss function is denoted by p and may
be obtained by solving the following equation
2
p = E( 2 |X)

The Bayesian estimator of the parameter under asymmetric Precautionary

loss function is,
1
2
((m
+
a
+
1)(m
+
a))
P =
(13)
b T (xi )
P ) and
The Bayesian estimator of the reliability function R(t) (denoted by R
E ) are,
hazard function H(t) (denoted by H

P =
R

m+a

b T (xi )

2et
b T (xi ) 2 log 1+e
t


P = P log 1 + e
H
2et

(14)

respectively.

Simulation

Applying the algorithm of [3], we used the following steps to generate a progressive Type II censored sample from the Rayleigh distribution:
(a) Simulate m independent exponential random variables Z1 , Z2 , ..., Zm . This
can be done using inverse transformation Zi = ln(1 Ui ) where Ui are independent unif orm(0, 1) random variables.
(b) Set
Xi =

Z2
Z3
Zi
Z1
+
+
+ +
n n R1 1 n R1 R2 2
n R1 R2 Ri i + 1

for i = 1, 2, ..., m. This is the required progressively type-II censored sample

from the standard exponential distribution.
(c) Finally, we set Yi = F 1 (1 exp(Xi )),for i = 1, 2, ..., m, where F 1 (.)
is the inverse cumulative distribution function of the generalized half logistic
distribution. Then Y1 , Y2, , Ym is the required progressively type-II censored
sample from the distribution F (.).

5258

R. Azimi et al

(d) Compute the MLE of by (5).

(e) Compute the the Bayes estimates S ,p ,E ,L respectively, using (7),
(9), (11), (13) and obtain dierent estimators for R(t) by using (8), (10), (12),
(14).
(d) Repeat the above steps 2000 times. We then obtain the means and the
RMSEs (root mean squared error) for dierent censoring sizes m and censoring
schemes where

RMSE =


2000


20001
2
( )
i=1

and is the estimator of .

In all above cases the prior parameters chosen as a = 0.5 and b = 1, which
yield the generated value of = 1 as the true value. The true values of R(t) in
t = 0.5 is obtained R(0.5) = 0.7550813. The results are summarized in Tables
1-2 (rst entries are estimators and second entries are RMSEs).

Conclusion

In this paper, we have presented the Bayesian estimates of the parameter ,

reliability R(t), and failure rate, H(t) , functions of the generalized half logistic distribution. According to Table-1 the Bayesian estimates of parameter
under LINEX loss function (when c = 0.5) have the smallest biased as compared with the estimates under squared-error loss function, Precautionary loss
function, entropy loss function, and maximum likelihood estimators (MLE).
However the Bayesian estimates under entropy loss function have the smallest estimated RMSE as compared with the estimates under squared-error loss
function, Precautionary loss function, LINEX loss function, and maximum
likelihood estimators (MLE). Since the Bayesian estimates under entropy loss
function have greater biased as compared with LINEX loss function we suggest to use LINEX loss function for estimating the parameter of Generalized
half logistic distribution.It is immediate to note that RMSEs decrease as n,
m increases. Table-2 shows that the Precautionary loss function is better
for estimating reliability function R(t) as compared with other loss functions
(squared-error loss function, entropy loss function, LINEX loss function) and
maximum likelihood method.

5259

Bayesian estimation of generalized half logistic distribution

Table 1: Averaged values of RMSEs for estimates of the parameter .

n
20

m
10

R = (R1 , ..., Rm )
(4,4,2,0,...,0)
(5,5,0,...,0)
(3,3,3,1,0,...,0)

15

(2,2,1,0,...,0)
(1,1,1,1,1,0,...,0)
(5,0,...,0)

30

15

(5,5,5,0,...,0)
(3,3,3,3,3,0..,0)
(4,4,4,3,0,...,0)

20

(5,5,0,...,0)
(3,3,3,1,0,...,0)
(2,2,2,2,2,0,...,0)

40

20

(5,5,5,5,0,...,0)
(3,3,3,3,3,3,2,0,...0)
(2,...,2,0,...,0)

30

(5,5,0,...,0)
(2,2,2,2,2,0,...,0)
(3,3,3,1,0,...,0)

1.1073
0.3990
1.1279
0.4224
1.1132
0.4098
1.0743
0.3143
1.0802
0.3065
1.0717
0.3084
1.0647
0.3078
1.0746
0.3096
1.0683
0.2920
1.0595
0.2605
1.0558
0.2598
1.0630
0.2506
1.0407
0.2500
1.0546
0.2558
1.0626
0.2548
1.0331
0.1965
1.0355
0.2002
1.0429
0.2093

S
1.1072
0.3152
1.0525
0.3311
1.0405
0.3211
1.0308
0.2704
1.0364
0.2634
1.0286
0.2665
1.0223
0.2660
1.0312
0.2668
1.0261
0.2524
1.0286
0.2329
1.0251
0.2334
1.0320
0.2242
1.0113
0.2262
1.0241
0.2296
1.0316
0.2284
1.0142
0.1838
1.0164
0.1868
1.0234
0.1948

L
1.0090
0.2963
1.0247
0.3099
1.0134
0.3011
1.0130
0.2591
1.0185
0.2521
1.0109
0.2557
1.0048
0.2555
1.0134
0.2558
1.0085
0.2422
1.0153
0.2253
1.0119
0.2262
1.0187
0.2168
1.0084
0.2200
1.0109
0.2225
1.0182
0.2209
1.0056
0.1801
1.0078
0.1830
1.0146
0.1905

E
0.9372
0.2902
0.9522
0.2996
0.9414
0.2940
0.9643
0.2538
0.9696
0.2459
0.9623
0.2507
0.9564
0.2518
0.9647
0.2504
0.9599
0.2382
0.9784
0.2209
0.9751
0.2221
0.9817
0.2119
0.9620
0.2183
0.9742
0.2187
0.9812
0.2159
0.9809
0.1782
0.9831
0.1808
0.9898
0.1873

P
1.0841
0.3384
0.9522
0.3569
1.0890
0.3450
1.0636
0.2844
1.0694
0.2779
1.0613
0.2802
1.0548
0.2790
1.0640
0.2808
1.0587
0.2656
1.0534
0.2426
1.0498
0.2428
1.0569
0.2343
1.0357
0.2341
1.0488
0.2389
1.0564
0.2384
1.0307
0.1887
1.0330
0.1920
1.0400
0.2005

5260

R. Azimi et al

Table 2: Averaged values of RMSEs for estimates of the reliability function

R(t).
n
20

m
10

R = (R1 , ..., Rm )
(4,4,2,0,...,0)
(5,5,0,...,0)
(3,3,3,1,0,...,0)

15

(2,2,1,0,...,0)
(1,1,1,1,1,0,...,0)
(5,0,...,0)

30

15

(5,5,5,0,...,0)
(3,3,3,3,3,0..,0)
(4,4,4,3,0,...,0)

20

(5,5,0,...,0)
(3,3,3,1,0,...,0)
(2,2,2,2,2,0,...,0)

40

20

(5,5,5,5,0,...,0)
(3,3,3,3,3,3,2,0,...0)
(2,...,2,0,...,0)

30

(5,5,0,...,0)
(2,2,2,2,2,0,...,0)
(3,3,3,1,0,...,0)

R
0.7367
0.0762
0.7328
0.0802
0.7356
0.0774
0.7420
0.0618
0.7407
0.0606
0.7425
0.0612
0.7440
0.0608
0.7419
0.0612
0.7430
0.0581
0.7443
0.0520
0.7451
0.0523
0.7435
0.0505
0.7482
0.0506
0.7453
0.0514
0.7436
0.0515
0.7491
0.0404
0.7487
0.0409
0.7472
0.0426

S
R
0.7533
0.0614
0.7502
0.0640
0.7525
0.0621
0.7526
0.0536
0.7514
0.0523
0.7531
0.0532
0.7544
0.0530
0.7525
0.0531
0.7534
0.0505
0.7521
0.0466
0.7528
0.0471
0.7512
0.0451
0.7556
0.0460
0.7530
0.0463
0.7514
0.0461
0.7540
0.0378
0.7536
0.0382
0.7522
0.0396

L
R
0.7680
0.0630
0.7648
0.0649
0.7671
0.0635
0.7677
0.0551
0.7664
0.0535
0.7681
0.0549
0.7694
0.0551
0.7676
0.0546
0.7685
0.0523
0.7674
0.0482
0.7681
0.0489
0.7666
0.0465
0.7709
0.0487
0.7683
0.0481
0.7667
0.0475
0.7696
0.0404
0.7691
0.0406
0.7677
0.0415

E
R
0.7471
0.0651
0.7437
0.0681
0.7461
0.0659
0.7485
0.0558
0.7472
0.0546
0.7489
0.0553
0.7502
0.0551
0.7483
0.0552
0.7492
0.0525
0.7490
0.04811
0.7497
0.0486
0.7481
0.0467
0.7526
0.0472
0.7499
0.0477
0.7483
0.0476
0.7520
0.0385
0.7515
0.0389
0.7501
0.0405

P
R
0.7563
0.0599
0.7532
0.0623
0.7554
0.0605
0.7547
0.0526
0.7534
0.0514
0.7551
0.0523
0.7563
0.0522
0.7545
0.0521
0.7554
0.0496
0.7536
0.0459
0.7543
0.0465
0.7528
0.0445
0.7571
0.0455
0.7545
0.0456
0.7529
0.0455
0.7550
0.0375
0.7546
0.0378
0.7532
0.0392

Bayesian estimation of generalized half logistic distribution

5261

References
[1] S.H. Arora, G.C. Bihmani and M.N. Patel, Some Results on Maximum
Likelihood Estimators of Parameters of Generalized Half Logistic Distribution under Type-I Progressive Censoring with Changing Failure Rate.
International Journal Of Contemporary Mathematical Sciences, 5(14)
(2010), 685-698.
[2] N. Balakrishnan, Progressive censoring methodology, an appraisal. TEST
16(2) (2007), 211-296 (with discussions)
[3] N. Balakrishnan and R. Aggarwala , Progressive censoring, theory, methods and applications. Birkhauser, Boston(2000)
[4] R. Calabria and G. Pulcini, An engineering approach to Bayes estimation
for the Weibull distribution, Micro-electron. Reliab. 34(5)(1994), 789-802.
[5] D.K. Dey and L.L. Pei-San, On comparison of estimators in a generalized
life model,Micro-electron. Reliab. , 32 (1)(1992), 207-221.
[6] D.K. Dey, M. Ghosh, and C. Srinivasan, Simultaneous estimation of parameters under entropy loss J. Statist. Plan. and Infer., (1987) 347-363.
[7] J.G. Norstrom, The use of precautionary loss functions in risk analysis,IEEE Transactions on Reliability, 45(3)(1996), 400-403.
[8] V. Ramakrsihnan, Generalizations to half logistic distribution and related
inference, PhD thesis, Acharya Nagarjuna University (AP), India, (2008).
Received: May, 2012